Overview
- TypeSafe launched Jev after two years in stealth with a $40 million seed round led by DCVC and is led by former OpenAI researcher Diogo Almeida, and the company removed the initial waitlist to open public access with $5 starter credits.
- Jev is a ‘System One’ decision model that returns typed, probabilistic primitives — Choice (one of many options), Score (a rubric value) and Noul (yes/no probability) — so software can use a structured answer and its calibrated probability instead of parsing freeform text.
- TypeSafe and early tests report typical call latencies around 100–150 milliseconds and a pricing model of $0.042 per million input tokens with free output tokens, and platform partners including Vercel, Netlify, Cloudflare, OpenRouter and LiteLLM have rapidly integrated the API.
- Developers and independent reviewers say Jev is strong out of the box for narrow classification and high‑cardinality choice sets but warn it must be deployed with shadow rollouts, full probability logging, empirically set thresholds and human or stronger‑model fallbacks because adversarial inputs and calibration gaps can steer decisions wrong.
- An open‑source rival, Laya, has appeared offering local, faster inference and fine‑tuning at the cost of out‑of‑the‑box calibration, and ongoing independent benchmarks are comparing hosted Jev’s immediate ease of use against self‑hosted tradeoffs in latency, privacy and customization.